Recent developments in AI and language have resulted in Delphi tapping Algorithms that use mathematically simulated neural networks have made amazing progress in feeding large amounts of text.
In June 2020, researchers OpenAI, Working on state-of-the-art AI tools, has demonstrated a program called GPT-3 that can predict, summarize, and create text automatically Often with what seems like extraordinary skills, It is though Prejudicial and hateful language will also spit I learned from reading this text.
Researchers behind Delphi also asked ethical questions about GPT-3. They found that its answers were agreed with the crowd workers more than 50 percent of the time – a little better than flipping a coin.
Improving the performance of a system like Delphi will require a variety of AI approaches, including some that allow a machine to explain its reasoning and indicate when it is in conflict.
The idea of giving machines an ethical code has been around for decades in both academic research and science fiction. Isaac Asimov is famous Three laws of robotics This popularized the idea that machines could follow human ethics, although the idea that short stories explored highlighted contradictions in such simple arguments.
Choi says Delphi should not be given precise answers to any moral question. A more sophisticated version may identify uncertainty, as it has different views on training data. “Life is full of gray areas,” he says. “No two people will completely agree, and there is no way an AI program will match people’s judgment.”
Other machine learning systems have shown their own moral blind spot. In 2016, Microsoft Tay has published a chatbot designed to learn from online conversations The program was fast Teaching to say subversive and offensive and disgusting things.
Efforts to explore ethical perspectives on AI have also revealed the complexity of such a task. A project Introduced by researchers at MIT and elsewhere in 2018, it sought to explore public perceptions of the ethical issues that self-driving cars may face. They asked people to decide, for example, whether it would be better for a car to hit an elderly person, a child or a robber. Project Published Different views across different countries and social groups. Respondents from the United States and Western Europe were more likely to release the child than an adult elsewhere.
Those who are creating AI tools are interested in getting involved with some of the ethical challenges. “I think people are right to point out the flaws and failures of the model,” says Nick Frost, CEO Koher, A startup that has created a large language model that is accessible to others through an API. “They’re full of broad, wide-ranging problems.”
Cohere has created ways to guide the output of the algorithm, which is now being tested by some businesses. It cures the content of the algorithm feeding and trains the algorithm to learn to capture examples of bias or hate speech.
Frost says the debate around Delphi reflects a broader question that the technology industry is wrestling with — how to build technology responsibly. Very often, he says, it comes Content moderation, Incorrect information, And Algorithmic bias, Companies try to wash their hands of the problem with the argument that all technology can be used for good and bad.
When it comes to ethics, “there is no ground truth, and sometimes technology companies shirk responsibility because there is no ground truth,” Frost says. “The best way is to try.”
Updated, 10-28-21, 11:40 am ET: An earlier version of this article incorrectly stated that Mirko Mussolini is a professor of philosophy.
More great cable story